US 20050192914 A1 Abstract A profile model can be selected for use in examining a structure formed on a semiconductor wafer using optical metrology by obtaining an initial profile model having a set of profile parameters. A machine learning system is trained using the initial profile model. A simulated diffraction signal is generated for an optimized profile model using the trained machine learning system, where the optimized profile model has a set of profile parameters with the same or fewer profile parameters than the initial profile model. A determination is made as to whether the one or more termination criteria are met. If the one or more termination criteria are met, the optimized profile model is modified and another simulated diffraction signal is generated using the same trained machine learning system.
Claims(50) 1. A method of selecting a profile model for use in examining a structure formed on a semiconductor wafer using optical metrology, the method comprising:
a) obtaining an initial profile model having a set of profile parameters; b) training a machine learning system using the initial profile model; c) generating a simulated diffraction signal for an optimized profile model using the trained machine learning system, wherein the optimized profile model has a set of profile parameters with the same or fewer profile parameters than the initial profile model; d) determining if one or more termination criteria are met; and e) if the one or more termination criteria are not met, modifying the optimized profile model and iterating steps c) to e), wherein the same trained machine learning system is used in iterating step c). 2. The method of obtaining a measured diffraction signal from an optical metrology device; and analyzing the simulated diffraction signal and the measured diffraction signal. 3. The method of 4. The method of 5. The method of obtaining a measured diffraction signal from an optical metrology device; and obtaining a profile associated with the measured diffraction signal, wherein the one or more termination criteria includes parameter correspondence determined between the profile parameters of the optimized profile model and dimensions of the profile associated with the measured diffraction signal. 6. The method of 7. The method of 8. The method of selecting at least one profile parameter of the optimized profile model to eliminate or fix to a value; and modifying the optimized profile model before iterating steps c) to e) by eliminating or fixing to a value the at least one profile parameter. 9. The method of training a first machine learning system using a set of training input data and a set of training output data,
wherein each of the training input data is a profile model having a set of profile parameters with the same profile parameters as the initial profile model, and
wherein each of the training output data is a diffraction signal.
10. The method of 11. The method of 12. The method of f) obtaining training input data; g) generating a diffraction signal with the first machine learning system using the training input data; h) determining if one or more termination criteria are met; and i) if the one or more termination criteria are not met, iterating steps g) to i). 13. The method of when iterating steps g) to i), adjusting the machine learning system or using new training input data in step g). 14. The method of testing the first machine learning system using a second machine learning system. 15. The method of before testing the first machine learning system, training the second machine learning system using the training input data for the first machine learning system as training output data for the second machine learning system, and the training output data for the first machine learning system as training input data for the second machine learning system. 16. The method of after training the second machine learning system,
generating one or more simulated diffraction signals using one or more profile models as inputs to the first machine learning system;
generating one or more profile models using the one or more simulated diffraction signals generated by the first machine learning system as inputs to the second machine learning system; and analyzing the one or more profile models generated by the second machine learning system and the one or more profile models used as inputs to the first machine learning system. 17. The method of 18. The method of 19. The method of 20. The method of if one or more termination criteria are met,
selecting at least one profile parameter of the optimized profile model; and
setting the at least one profile parameter to a determined value.
21. The method of 22. A computer-readable storage medium containing computer executable instructions for causing a computer to select a profile model for use in examining a structure formed on a semiconductor wafer using optical metrology, comprising instructions for:
a) obtaining an initial profile model having a set of profile parameters; b) training a machine learning system using the initial profile model; c) generating a simulated diffraction signal for an optimized profile model using the trained machine learning system, wherein the optimized profile model has a set of profile parameters with the same or fewer profile parameters than the initial profile model; d) determining if one or more termination criteria are met; and e) if the one or more termination criteria are not met, modifying the optimized profile model and iterating steps c) to e), wherein the same trained machine learning system is used in iterating step c). 23. The computer-readable storage medium of obtaining a measured diffraction signal from an optical metrology device; and analyzing the simulated diffraction signal and the measured diffraction signal. 24. The computer-readable storage medium of 25. The computer-readable storage medium of 26. The computer-readable storage medium of obtaining a measured diffraction signal from an optical metrology device; and obtaining a profile associated with the measured diffraction signal, wherein the one or more termination criteria includes parameter correspondence determined between the profile parameters of the optimized profile model and dimensions of the profile associated with the measured diffraction signal. 27. The computer-readable storage medium of 28. The computer-readable storage medium of 29. The computer-readable storage medium of selecting at least one profile parameter of the optimized profile model to eliminate or fix to a value; and modifying the optimized profile model before iterating steps c) to e) by eliminating or fixing to a value the at least one profile parameter. 30. The computer-readable storage medium of training a first machine learning system using a set of training input data and a set of training output data,
wherein each of the training input data is a profile model having a set of profile parameters with the same profile parameters as the initial profile model, and
wherein each of the training output data is a diffraction signal.
31. The computer-readable storage medium of 32. The computer-readable storage medium of testing the first machine learning system using a second machine learning system. 33. The computer-readable storage medium of before testing the first machine learning system, training the second machine learning system using the training input data for the first machine learning system as training output data for the second machine learning system and the training output data for the first machine learning system as training input data for the second machine learning system. 34. The computer-readable storage medium of after training the second machine learning system,
generating one or more simulated diffraction signals using one or more profile models as inputs to the first machine learning system;
generating one or more profile models using the one or more simulated diffraction signals generated by the first machine learning system as inputs to the second machine learning system; and
analyzing the one or more profile models generated by the second machine learning system with the one or more profile models used as inputs to the first machine learning system.
35. The computer-readable storage medium of if one or more termination criteria are met,
selecting at least one profile parameter of the optimized profile model; and
setting the at least one profile parameter to a determined value.
36. The computer-readable storage medium of 37. A system to select a profile model for use in examining a structure formed on a semiconductor wafer using optical metrology, the system comprising:
an optical metrology device configured to provide a measured diffraction signal; a first machine learning system trained using an initial profile model having a set of profile parameters, the first machine learning system configured to generate a simulated diffraction signal for an optimized profile model having a set of profile parameters with the same or fewer profile parameters than the initial profile model, wherein if one or more termination criteria are not met, the optimized profile model is modified and the first machine learning system generates another simulated diffraction signal. 38. The system of 39. The system of 40. The system of 41. The system of 42. The system of 43. The system of 44. The system of a second machine learning system trained using the training input data for the first machine learning system as training output data for the second machine learning system, and the training output data for the first machine learning system as training input data for the second machine learning system. 45. The system of one or more simulated diffraction signals are generated using one or more profile models as inputs to the first machine learning system; one or more profile models are generated using the one or more simulated diffraction signals generated by the first machine learning system as inputs to the second machine learning system; and the one or more profile models generated by the second machine learning system are compared with the one or more profile models used as inputs to the first machine learning system. 46. The system of 47. The system of 48. The system of 49. The system of 50. The system of Description 1. Field of the Invention The present application relates to metrology of structures formed on semiconductor wafers, and more particularly to selecting a profile model for use in optical metrology using a machine learning system. 2. Related Art Optical metrology involves directing an incident beam at a structure, measuring the resulting diffracted beam, and analyzing the diffracted beam to determine a feature of the structure. In semiconductor manufacturing, optical metrology is typically used for quality assurance. For example, after fabricating a periodic grating in proximity to a semiconductor chip on a semiconductor wafer, an optical metrology system is used to determine the profile of the periodic grating. By determining the profile of the periodic grating, the quality of the fabrication process utilized to form the periodic grating, and by extension the semiconductor chip proximate the periodic grating, can be evaluated. One conventional optical metrology system uses a diffraction modeling technique, such as rigorous coupled wave analysis (RCWA), to analyze the diffracted beam. More particularly, in the diffraction modeling technique, a model diffraction signal is calculated based, in part, on solving Maxwell's equations. Calculating the model diffraction signal involves performing a large number of complex calculations, which can be time consuming and costly. In one exemplary embodiment, a profile model can be selected for use in examining a structure formed on a semiconductor wafer using optical metrology by obtaining an initial profile model having a set of profile parameters. A machine learning system is trained using the initial profile model. A simulated diffraction signal is generated for an optimized profile model using the trained machine learning system, where the optimized profile model has a set of profile parameters with the same or fewer profile parameters than the initial profile model. A determination is made as to whether the one or more termination criteria are met. If the one or more termination criteria are met, the optimized profile model is modified and another simulated diffraction signal is generated using the same trained machine learning system. The present invention can be best understood by reference to the following description taken in conjunction with the accompanying drawing figures, in which like parts may be referred to by like numerals: The following description sets forth numerous specific configurations, parameters, and the like. It should be recognized, however, that such description is not intended as a limitation on the scope of the present invention, but is instead provided as a description of exemplary embodiments. 1. Optical Metrology With reference to As depicted in Optical metrology system 2. Library-Based Process In a library-based process, the measured diffraction signal is compared to a library of simulated diffraction signals. More specifically, each simulated diffraction signal in the library is associated with a profile model of the structure. When a match is made between the measured diffraction signal and one of the simulated diffraction signals in the library or when the difference of the measured diffraction signal and one of the simulated diffraction signals in the library is within a preset or matching criterion, the profile model associated with the matching simulated diffraction signal in the library is presumed to represent the actual profile of the structure. A feature of the structure can then be determined based on the profile model associated with the matching simulated diffraction signal. Thus, with reference again to The set of profile models stored in library For example, as depicted in As described above, the set of profile models stored in library Thus, the profile parameters of the profile model associated with a matching simulated diffraction signal can be used to determine a feature of the structure being examined. For example, a profile parameter of the profile model corresponding to a bottom CD can be used to determine the bottom CD of the structure being examined. With reference again to For a more detailed description of a library-based process, see U.S. patent application Ser. No. 09/907,488, titled GENERATION OF A LIBRARY OF PERIODIC GRATING DIFFRACTION SIGNALS, filed on Jul. 16, 2001, which is incorporated herein by reference in its entirety. 3. Regression-Based Process In a regression-based process, the measured diffraction signal is compared to a simulated diffraction signal generated prior to the comparison (i.e., a trial simulated diffraction signal) using a set of profile parameters (i.e., trial profile parameters) for a profile model. If the measured diffraction signal and the trial simulated diffraction signal do not match or when the difference of the measured diffraction signal and the trial simulated diffraction signal is not within a preset or matching criterion, another trial simulated diffraction signal is generated using another set of profile parameters for another profile model, then the measured diffraction signal and the newly generated trial simulated diffraction signal are compared. When the measured diffraction signal and the trial simulated diffraction signal match or when the difference of the measured diffraction signal and the trial simulated diffraction signals is within a preset or matching criterion, the profile model associated with the matching trial simulated diffraction signal is presumed to represent the actual profile of the structure. The profile model associated with the matching trial simulated diffraction signal can then be used to determine a feature of the structure being examined. Thus, with reference again to In one exemplary embodiment, the trial simulated diffraction signals and profile models can be stored in a library For a more detailed description of a regression-based process, see U.S. patent application Ser. No. 09/923,578, titled METHOD AND SYSTEM OF DYNAMIC LEARNING THROUGH A REGRESSION-BASED LIBRARY GENERATION PROCESS, filed on Aug. 6, 2001, which is incorporated herein by reference in its entirety. 4. Selecting Optimal Profile Model The accuracy, complexity, and length of time needed to perform a library-based process and/or regression-based process depends, in part, on the complexity of the profile model used. For example, increasing the complexity of the profile model by adding a profile parameter can increase accuracy. However, the increased complexity of the profile model can increase the complexity and the amount of time needed to perform the library-based process and/or regression-based process. Thus, with reference to In step In step In step With reference to As depicted in In neural network In neural network With reference to In step In step In step A cost function determined between two diffraction signals is illustrated by the equations below, where V A goodness of fit (GOF) is a measure of the proximity of two sets values. For example, when ellipsometric measurements are used, GOF can be determined based on values for tan ψ and cos Δ, where tan ψ and cos Δ are represented by a single vector of n dimensions:
One commonly used formula for GOF between a first signal S A correlation coefficient, r, between two profile parameters can be calculated using the formula:
A sensitivity of a diffraction signal to changes in one or more profile parameters can be determined by changing one profile parameter by a small amount and keeping the other profile parameters constant. For example, the sensitivity of profile parameter x The sensitivity of a profile parameter may be quantitatively expressed by calculating the sum-square-error (SSE) of the changed diffraction signal compared to the diffraction signal using nominal values. The SSE formula is as follows:
A confidence interval of a profile parameter can be determined by the amount of change from a nominal value of the profile parameter, where the change in the diffraction signals is greater than the noise level. The noise in the diffraction signals may be due to system noise, for example, noise from the measurement devices, or the noise may be simulated. The confidence interval is generally expressed as a multiple of the standard deviation sigma, σ, of the profile parameter. The standard deviation for a profile parameter can be calculated from measured values of the profile parameter, using the formula:
The confidence interval is typically calculated from a given set of sample input data representing actual measurements off the wafer structure. The confidence interval may also be calculated using simulated random noise introduced in the measurement data for the profile parameter. As depicted in It should be recognized that training process Furthermore, training process With reference to In In In With reference to In step After the second machine learning system has been trained, in step In step An empirical risk minimization (ERM) technique can be used to quantify how well the trained machine learning system can generalize to new input. For a more detailed description of ERM, see “Statistical Learning Theory” by Vladimir N. Vapnik, Wiley-Interscience, September 1998, which is incorporated herein by reference in its entirety. With reference again to In step In the present exemplary embodiment, when cost function is included as a termination criterion, a cost function value can be determined between the simulated diffraction signal and the measured diffraction signal. The determined cost function can then be compared to a preset cost function value to determine if the determined cost function value is less than or equal to the preset cost function value. The preset cost function value may be set at a specific number, for example, 0.05. When GOF value is included as a termination criterion, a GOF value can be determined between the simulated diffraction signal and the measured diffraction signal. The determined GOF value can then be compared to a preset GOF value to determine if the determined GOF value is less than or equal to the preset GOF value. The preset GOF value may be set at a specific number, for example 0.95. When correspondence is included as a termination criterion, a correspondence is obtained between the profile parameters of the optimized profile model and the dimensions of the actual profile that corresponds to the measured diffraction signal. In the present exemplary embodiment, the dimensions of the actual profile can be obtained using SEM. When correlation is included as a termination criterion, a correlation coefficient can be determined between a pair of profile parameters of the optimized profile model. The determined correlation coefficient can then be compared to a preset correlation coefficient to determine if the determined correlation coefficient is less than or equal to the preset correlation coefficient. When parameter sensitivity is included as a termination criterion, a sensitivity can be determined for each profile parameter of the optimized profile model. The determined sensitivity can then be compared to a preset sensitivity to determine if the determined sensitivity is less than or equal to the preset sensitivity coefficient. When confidence interval is included as a termination criterion, a confidence interval is determined for each profile parameter of the optimized profile model. The determined confidence interval can then be compared to a preset confidence interval to determine if the determined confidence interval is less than or equal to the preset confidence interval. The preset confidence interval may be set to any number of sigma, such as three-sigma. In step For example, with reference to With reference again to For additional examples of profile model selection processes, see U.S. patent application Ser. No. 10/206,491, titled MODEL AND PARAMETER SELECTION FOR OPTICAL METROLOGY, filed on Jul. 25, 2002, which is incorporated herein by reference in its entirety. See also U.S. patent application Ser. No. 10/397,631, titled OPTIMIZED MODEL AND PARAMETER SELECTION FOR OPTICAL METROLOGY, filed on Mar. 25, With reference again to In the exemplary embodiment, after selecting an optimized profile model using process For example, assume a selection criteria includes a correlation of at least 0.95 or higher. Now assume that an optimized profile model includes a width parameter and a thickness parameter with a correlation greater than 0.95. Thus, in this example, the width parameter and/or the thickness parameter is selected and set to a determined value. Assume that the thickness parameter in the example above is selected. Now assume that the determined value is obtained using an averaging technique. More particularly, in the present example, multiple thickness measurements of the selected thickness parameter on a wafer are obtained. An average thickness measurement of the selected thickness parameter is then calculated from the multiple thickness measurements. The selected thickness parameter is then set to the average thickness measurement. Note that when a profile refinement process is used with a machine learning system, a selected profile parameter can be set to any value. However, when a profile process is used with a library-based system, a selected profile parameter is preferably set based on a constraint of the library, such as the resolution of the library. For example, if a profile process is used with a machine learning system and an average thickness measurement is 50.25 nanometers, then the selected thickness parameter can be set to 50.25 nanometers. However, if a profile process is used with a library-based system and the library includes thickness parameters at intervals of 50, 55, and 60 nanometers, then the selected thickness parameter is set to 50 nanometers. For a more detailed description of a profile refinement process, see U.S. patent application Ser. No. 10/735,212, titled PARAMETRIC OPTIMIZATION OF OPTICAL METROLOGY MODEL, filed on Dec. 12, 2003, which is incorporated herein by reference in its entirety. Although exemplary embodiments have been described, various modifications can be made without departing from the spirit and/or scope of the present invention. Therefore, the present invention should not be construed as being limited to the specific forms shown in the drawings and described above. Referenced by
Classifications
Legal Events
Rotate |